-
KIP-932 adds a long awaited capability to the Apache Kafka project: queue-like semantics, including the ability to acknowledge messages on a one-by-one basis. This positions Kafka for use cases such as job queuing, for which it hasn’t been a good fit historically. As multiple members of a share group can process the messages from a single topic partition, the partition count does not limit the degree of consumer parallelism any longer. The number of consumers in a group can quickly be increased and decreased as needed, without requiring to repartition the topic.
[….] Available as an early access feature as of the [unreleased] Kafka 4.0 release, Kafka queues are not recommended for production usage yet, and there are several limitations worth calling out: most importantly, the lack of DLQ support. More control over retry timing would be desirable, too. As such, I don’t think Kafka queues in their current form will make users of established queue solutions such as Artemis or RabbitMQ migrate to Kafka. It is a very useful addition to the Kafka feature set nevertheless, coming in handy for instance for teams already running Kafka and who look for a solution for simple queuing use cases, avoiding to stand up and operate a separate solution just for these. This story will become even more compelling if the feature gets built out and improved in future Kafka releases.
Tags: kafka queueing queues architecture